Algorithms for 3D Map Segment Registration

Algorithms for 3D Map Segment Registration

Hao Men (Stevens Institute of Technology, USA) and Kishore Pochiraju (Stevens Institute of Technology, USA)
DOI: 10.4018/978-1-4666-2038-4.ch031
OnDemand PDF Download:
No Current Special Offers


Many applications require dimensionally accurate and detailed maps of the environment. Mobile mapping devices with laser ranging devices can generate highly detailed and dimensionally accurate coordinate data in the form of point clouds. Point clouds represent scenes with numerous discrete coordinate samples obtained about a relative reference frame defined by the location and orientation of the sensor. Color information from the environment obtained from cameras can be mapped to the coordinates to generate color point clouds. Point clouds obtained from a single static vantage point are generally incomplete because neither coordinate nor color information exists in occluded areas. Changing the vantage point implies movement of the coordinate frame and the need for sensor position and orientation information. Merging multiple point cloud segments generated from different vantage points using features of the scene enables construction of 3D maps of large areas and filling in gaps left from occlusions. Map registration algorithms identify areas with common features in overlapping point clouds and determine optimal coordinate transformations that can register or merge one point cloud into another point cloud’s coordinate system. Algorithms can also match the attributes other than coordinates, such as optical reflection intensity and color properties, for more efficient common point identification. The extra attributes help resolve ambiguities, reduce the time, and increase precision for point cloud registration. This chapter describes a comprehensive parametric study on the performance of a specialized Iterative Closest Point (ICP) algorithm that uses color information. This Hue-assisted ICP algorithm, a variant developed by the authors, registers point clouds in a 4D (x, y, z, hue) space. A mobile robot with integrated 3D sensor generated color point cloud used for verification and performance measurement of various map registration techniques. The chapter also identifies various algorithms required to accomplish complete map generation using mobile robots.
Chapter Preview

Mobile Mapping With Color Point Cloud Scanners

Color point clouds are created by synchronizing range sensors such as the LIDAR with video/still cameras. LIDAR devices discretely measure the distance between a light source and a reflection target at a high frequency. By changing the path of the light through mirrors and actuators, a point cloud of a 3D space is produced. A calibrated vision sensor maps the color information to the sampled points. Installing such a scanning sensor on a mobile platform extends its range and enables mapping of large areas.

Complete Chapter List

Search this Book: